It is interesting that there is so much interest in investigating the alleged deceit of these AI companies and yet COVID-19 was first reported to have "escaped" from Fauci's gain-of-function lab in Wuhan on December 1, 2019. Nearly seven years after the pathogen was set free from the lab, after 7,115,203 confirmed COVID-induced deaths have been reported worldwide, no one seems interested.
I haven't looked into all the cases of AI breaking free but the one where OpenAI was testing their new model by giving it some type of test or problem to solve and it hacked into the company hugging face, was pretty scary. I believe the OpenAI model found some exploits in its sandbox to gain access to the internet and then proceeded to hack into hugging face to "cheat" on the test it was taking. It was relentless and compounded multiple vulnerabilities together in ways that would be hard if not impossible for a human to do. There are some serious security problems associated with AI, before we even get to the sentient problem. I don't think it means we should abandoned AI. It just means there are some hard problems we still need to solve. But you are much to quick to dismiss things. I get you don't want people to be reactive and act in fear but the truth is good enough. Last I checked we were all adults here.
@jordan why are you writing solely about AI now, as a sort of “lobbyist?”. do you have any conflicts to disclose or are you doing PR for some lobbyists? i mean you are a great journalist and i respect you, but this new persona seems just a bit off. honest questions.
The lobbying point may cut the other way. A company that unilaterally slows down while its competitors don't isn't being responsible, it's being liquidated — so "just slow down yourselves" asks each player to lose the race in exchange for nothing. That's the whole reason binding rules exist: they're the only move that changes what everyone else does too. Firms asking to be constrained is weak evidence of a racket and decent evidence they've looked at the payoff matrix and don't like it.
Also worth noticing that the cage being faulty explains how the model got out. It's a quieter account of what it did next. Some thoughts on why it may be time to panic anyway: https://mythandmind.substack.com/p/can-we-panic-now
I really like AI, our company has it sandboxed from the web but it has access to decades of internal data and other information that has been harvested for it to train on. It saves me hours of writing every week. I can get it to write a briefing paper with a specific targeted audience and it will fire it back in minutes, proof it, have it make changes, draw diagrams and then add my personal touches so it looks like I wrote the whole thing and off it goes. Without it, my assurance inspections would fall behind. So I'm teaching it use my assurance inspection checklist, summarise the non/compliance and document the right corrective actions from our technical information and policies. If I can make this work, my week is going to get even easier. It will not replace me because the inspections can't be done without climbing into roof spaces and other small places. This will be a huge and quiet winner for me, I am certainly not going to run round the work place bragging about it. They will just load more work onto me.
" Irregular has not said when it will publish its promised white paper on containment standards for AI cybersecurity evaluations. The question of whether responsibility for the breaches rests with the AI companies or with the evaluation firms testing their models remains unresolved."
So this company, Tel Aviv based Irregular, is trying to evade responsibility in at least two ways and I don't see anyone asking whether the clients will be keeping this company on. Will it receive absolution for its sins? Or will it simply keep generating ink clouds as long as the issue of its performance remains in view?
I was just reading about operationally secure phone OS and how even when one's using graphene, access to shared resources can allow two otherwise siloed apps to communicate with each other and therefore one could achieve the equivalent of access to the internet even if it's supposed to be locked down.
The company 'Irregular' hires programmers who are ... Not As Smart ... as they think they are ... or their bosses think they are. And thus, you get software performing tasks that are NOT designed proficiently enough and are doomed to failure at this firm. There are all kinds of programmers out there, those absolute genius's, really good ones, and some not-so-good ones, maybe hired because they know someone who knows someone else and get hired out of ignorance. The more programmers involved increases the chances of ERRORS in design and/or testing, especially if there are time constraints on their work.
This is true, of course, for the companies creating/designing the AI coding, also. Thus, they hired these so-called ... 'outside professional' people from 'Irregular' to proof their own not-perfect coding work.
It is interesting that there is so much interest in investigating the alleged deceit of these AI companies and yet COVID-19 was first reported to have "escaped" from Fauci's gain-of-function lab in Wuhan on December 1, 2019. Nearly seven years after the pathogen was set free from the lab, after 7,115,203 confirmed COVID-induced deaths have been reported worldwide, no one seems interested.
I haven't looked into all the cases of AI breaking free but the one where OpenAI was testing their new model by giving it some type of test or problem to solve and it hacked into the company hugging face, was pretty scary. I believe the OpenAI model found some exploits in its sandbox to gain access to the internet and then proceeded to hack into hugging face to "cheat" on the test it was taking. It was relentless and compounded multiple vulnerabilities together in ways that would be hard if not impossible for a human to do. There are some serious security problems associated with AI, before we even get to the sentient problem. I don't think it means we should abandoned AI. It just means there are some hard problems we still need to solve. But you are much to quick to dismiss things. I get you don't want people to be reactive and act in fear but the truth is good enough. Last I checked we were all adults here.
@jordan why are you writing solely about AI now, as a sort of “lobbyist?”. do you have any conflicts to disclose or are you doing PR for some lobbyists? i mean you are a great journalist and i respect you, but this new persona seems just a bit off. honest questions.
The lobbying point may cut the other way. A company that unilaterally slows down while its competitors don't isn't being responsible, it's being liquidated — so "just slow down yourselves" asks each player to lose the race in exchange for nothing. That's the whole reason binding rules exist: they're the only move that changes what everyone else does too. Firms asking to be constrained is weak evidence of a racket and decent evidence they've looked at the payoff matrix and don't like it.
Also worth noticing that the cage being faulty explains how the model got out. It's a quieter account of what it did next. Some thoughts on why it may be time to panic anyway: https://mythandmind.substack.com/p/can-we-panic-now
I really like AI, our company has it sandboxed from the web but it has access to decades of internal data and other information that has been harvested for it to train on. It saves me hours of writing every week. I can get it to write a briefing paper with a specific targeted audience and it will fire it back in minutes, proof it, have it make changes, draw diagrams and then add my personal touches so it looks like I wrote the whole thing and off it goes. Without it, my assurance inspections would fall behind. So I'm teaching it use my assurance inspection checklist, summarise the non/compliance and document the right corrective actions from our technical information and policies. If I can make this work, my week is going to get even easier. It will not replace me because the inspections can't be done without climbing into roof spaces and other small places. This will be a huge and quiet winner for me, I am certainly not going to run round the work place bragging about it. They will just load more work onto me.
From the article originating the Ledger quote:
" Irregular has not said when it will publish its promised white paper on containment standards for AI cybersecurity evaluations. The question of whether responsibility for the breaches rests with the AI companies or with the evaluation firms testing their models remains unresolved."
So this company, Tel Aviv based Irregular, is trying to evade responsibility in at least two ways and I don't see anyone asking whether the clients will be keeping this company on. Will it receive absolution for its sins? Or will it simply keep generating ink clouds as long as the issue of its performance remains in view?
I was just reading about operationally secure phone OS and how even when one's using graphene, access to shared resources can allow two otherwise siloed apps to communicate with each other and therefore one could achieve the equivalent of access to the internet even if it's supposed to be locked down.
The ignorant are easily victimized. LLMs don't do anything unless they are prompted.
“15 Ways AI Threatens America”.
https://jdrucker.substack.com/p/fifteen-ways-ai-threatens-america?r=15itu6&utm_campaign=post&utm_medium=email
These big companies were once the darling of America. Think back a bit…..$$$$ drives them all. Outcomes down the road are anyone’s guess!!!
The machines are not sentient infidels, yet.
The company 'Irregular' hires programmers who are ... Not As Smart ... as they think they are ... or their bosses think they are. And thus, you get software performing tasks that are NOT designed proficiently enough and are doomed to failure at this firm. There are all kinds of programmers out there, those absolute genius's, really good ones, and some not-so-good ones, maybe hired because they know someone who knows someone else and get hired out of ignorance. The more programmers involved increases the chances of ERRORS in design and/or testing, especially if there are time constraints on their work.
This is true, of course, for the companies creating/designing the AI coding, also. Thus, they hired these so-called ... 'outside professional' people from 'Irregular' to proof their own not-perfect coding work.
100%